Predicting need for hospital-specific interventional care after surgery using electronic health record data

Area of Science:

• Surgical Oncology
• Health Services Research
• Predictive Analytics

Background:

• Prolonged hospital stays for surgical inpatients are common.
• Early identification of patients not needing hospital-specific care can facilitate timely discharge.
• Predicting the need for continued hospital care beyond the second postoperative day is crucial for efficient patient management.

Purpose of the Study:

• To develop and validate a predictive model for identifying surgical patients requiring hospital-specific interventional care beyond postoperative day 2.
• To improve patient flow and hospital capacity management through early discharge planning.

Main Methods:

• Retrospective study of adult surgical oncology patients (June 2017-February 2020).
• Defined hospital-specific care as unplanned reoperations, radiological interventions, or intravenous antibiotics.
• Compared analytical methods using AUC, sensitivity, specificity, PPV, and NPV.

Main Results:

• The study included 1,174 patient episodes; 50.5% required intervention.
• A random forest model achieved an AUC of 0.88 (95% CI 0.83-0.93).
• The model demonstrated 79.1% sensitivity and 80.0% specificity for predicting the need for continued hospital care.

Conclusions:

• A random forest model effectively predicts patients suitable for discharge to a nursing home, avoiding unnecessary hospital care.
• This predictive tool can help address hospital capacity challenges and enhance patient flow.
• Implementing such a model supports timely access to surgical care.